A machine learning-based system and method for identifying digital threats that includes implementing a machine learning-based digital threat mitigation service over a distributed network of computers; constructing, by the machine learning-based digital threat mitigation service, a subscriber-specific machine learning ensemble that includes a plurality of distinct machine learning models, wherein each of the plurality of distinct machine learning models is configured to perform a distinct machine learning task for identifying a digital threat or digital fraud; constructing a corpus of subscriber-specific digital activity data for training the plurality of distinct machine learning models of the subscriber-specific ensemble; training the subscriber-specific ensemble using at least the corpus of subscriber-specific digital activity data; and deploying the subscriber-specific ensemble.
Legal claims defining the scope of protection, as filed with the USPTO.
1. A machine learning-based system for identifying digital threats, the system comprising: a distributed network of computers implementing a machine learning-based digital threat mitigation service that: constructs a subscriber-specific machine learning ensemble of a plurality of distinct machine learning models, wherein each of the plurality of distinct machine learning models is configured to perform a distinct machine learning task for identifying a digital threat or digital fraud; constructs a corpus of subscriber-specific digital activity data for training the plurality of distinct machine learning models of the subscriber-specific machine learning ensemble; trains the subscriber-specific machine learning ensemble using at least the corpus of subscriber-specific digital activity data; identifies weights of a component mixture of the subscriber-specific machine learning ensemble during a validation phase of the subscriber-specific machine learning ensemble, wherein the validation phase of the subscriber-specific machine learning ensemble is performed using a prescribed subset of the corpus of subscriber-specific digital activity data, wherein the validation phase includes: generating by the subscriber-specific machine learning ensemble a plurality of threat scores based on an input of the prescribed subset of the corpus of subscriber-specific digital activity data; generating by a distinct global machine learning model a plurality of global threat scores based on an input of the prescribed subset of the corpus of subscriber-specific digital activity data; and measuring the plurality of threat scores produced by the subscriber-specific machine learning ensemble against the plurality of global threat scores produced by the global machine learning model; deploys the subscriber-specific machine learning ensemble.
2. The system according to claim 1 , wherein the machine learning-based digital threat mitigation service: calibrates threat scores generated by the subscriber-specific machine learning ensemble against threat scores generated by a pre-existing, global machine learning model; and migrates the subscriber from the pre-existing, global machine learning model to the subscriber-specific machine learning ensemble based on the calibration.
3. The system according to claim 1 , wherein: deploying the subscriber-specific machine learning ensemble is based on a successful calibration of threat scores produced by the subscriber-specific machine learning ensemble; and deploying the subscriber-specific machine learning ensemble includes exposing the threat scores of the subscriber-specific machine learning ensemble in lieu of threat scores produced by a pre-existing, global machine learning model that is implemented by the machine learning-based digital threat mitigation service.
4. The system according to claim 1 , wherein subscriber-specific digital activity data includes data relating to a plurality of transactions performed by online users or the like using one or more online resources and/or services of the subscriber to the digital threat scoring service.
5. The system according to claim 1 , wherein the subscriber-specific machine learning ensemble comprises a machine learning-based digital threat scoring model that is associated with an account of the subscriber hosted by the machine learning-based digital threat mitigation service.
6. The system according to claim 1 , wherein a composition of the plurality of distinct machine learning models defining the subscriber-specific machine learning ensemble is based on one or more attributes of the corpus of subscriber-specific digital activity data.
7. The system according to claim 1 , wherein the plurality of distinct machine learning models defining the subscriber-specific machine learning ensemble include a subset distinct machine learning models selected from a plurality of machine learning models made available by the machine learning-based digital threat mitigation service.
8. The system according to claim 1 , wherein the machine learning-based digital threat mitigation service: sets a training and/or validation configuration of the corpus of subscriber-specific digital activity data, wherein setting the training and/or validation configuration includes configuring the corpus of subscriber-specific digital activity data to make available only a first subset of the corpus of subscriber-specific digital activity data during a first phase of training the subscriber-specific machine learning ensemble.
9. The system according to claim 1 , wherein the machine learning-based digital threat mitigation service: sets a training and/or validation configuration of the corpus of subscriber-specific digital activity data, wherein setting the training and/or validation configuration includes configuring the corpus of subscriber-specific digital activity data to make available only a second subset of the corpus of subscriber-specific digital activity data during a second phase of validating the subscriber-specific machine learning ensemble.
10. The system according to claim 1 , wherein: the machine learning-based digital threat mitigation service to a plurality of distinct subscribers; and the subscriber-specific machine learning ensemble is distinct from other subscriber ensembles associated with other subscribers of the plurality distinct subscribers.
11. The system according to claim 1 , wherein identifying weights includes: computing a linear set of weights for each of the plurality of distinct machine learning models defining the component mixture of the subscriber-specific machine learning ensemble.
12. A machine learning-based method for identifying digital threats, the method comprising: implementing a machine learning-based digital threat mitigation service over a distributed network of computers; constructing, by the machine learning-based digital threat mitigation service, a subscriber-specific machine learning ensemble that includes a plurality of distinct machine learning models, wherein each of the plurality of distinct machine learning models is configured to perform a distinct machine learning task for identifying a digital threat or digital fraud; constructing a corpus of subscriber-specific digital activity data for training the plurality of distinct machine learning models of the subscriber-specific ensemble; training the subscriber-specific ensemble using at least the corpus of subscriber-specific digital activity data; identifying weights of a component mixture of the subscriber-specific machine learning ensemble during a validation phase of the subscriber-specific machine learning ensemble, wherein the validation phase of the subscriber-specific machine learning ensemble is performed using a prescribed subset of the corpus of subscriber-specific digital activity data, wherein the validation phase includes: generating by the subscriber-specific machine learning ensemble a plurality of threat scores based on an input of the prescribed subset of the corpus of subscriber-specific digital activity data; generating by a distinct global machine learning model a plurality of global threat scores based on an input of the prescribed subset of the corpus of subscriber-specific digital activity data; and measuring the plurality of threat scores produced by the subscriber-specific machine learning ensemble against the plurality of global threat scores produced by the global machine learning model; and deploying the subscriber-specific ensemble.
13. The method according to claim 12 , wherein the subscriber-specific machine learning ensemble comprises a machine learning-based digital threat scoring model that is associated with an account of the subscriber hosted by the machine learning-based digital threat mitigation service.
14. The method according to claim 12 , wherein identifying weights includes: computing a linear set of weights for each of the plurality of distinct machine learning models defining the component mixture of the subscriber-specific machine learning ensemble.
15. The method according to claim 12 , wherein identifying weights includes: computing a non-linear set of weights for each of the plurality of distinct machine learning models defining the component mixture of the subscriber-specific machine learning ensemble.
Cooperative Patent Classification codes for this invention. Click any code to explore related patents in that topic.
May 31, 2019
November 26, 2019
Browse 5M+ US patents with plain-English claim translations and AI-generated analysis.